Toward Robust Estimation of the Components of Forest Population Change
Abstract
Multiple levels of simulation are used to test the robustness of estimators of the components of change. I first created a variety of spatial-temporal populations based on, but more variable than, an actual forest monitoring data set and then sampled those populations under a variety of sampling error structures. The performance of each of four estimation approaches is evaluated when the temporal scale of the estimand of interest is 1 year while the temporal scale of observation is t years. Three approaches for estimating the individual components of forest change are compared over five simulated populations under four sets of sampling error structure. The performance of a modification to these approaches is shown when extraneously obtained information indicates that a deviation to the assumed population model exists. Finally, the extraneous information is incorporated into a mixed estimator, combining each of three general transition models with a single compatibility model. The first three approaches, without the incorporation of extraneous information, are compatible with large monitoring efforts that require intervention-free results. The mixed-estimation approach accounts for model assumptions that sometimes remain latent in other approaches and is amenable to the incorporation of the extraneously obtained information and to ensuring estimator compatibility. All four approaches are shown to work well when the sampling error structure is unbiased, while some notable differences in performance were observed at the temporal extremities of observation in the presence of temporal anomalies and in the presence of biased sampling error structures.